Is Ubuntu really going low-spec?
We all want to run Linux on 64 MB of RAM. But Ubuntu Lite’s way is not the right one.
:: Reviews : Ubuntu Goes Low Spec! (found here) introduced me to the new project called Ubuntu Lite.
What’s Ubuntu Lite? It’s a new Linux distribution. Its goal is to make a usable Linux-based system that works on 128 MB of RAM. How they plan to achieve this goal seems straightforward: around “lightweight” applications.
And it’s is exactly the wrong way to build a low-footprint distribution. Why?
Saving pennies, squandering dollars
First, the applications and building blocks chosen are simpler. In other words, they have less functionality than the two standard behemoths: KDE and GNOME.
Say it out loud: they’re building a cripple from the outset. It will never be as functional as any environment that comes with either KDE or GNOME.
Does that lead to a low memory footprint?
Second, the project is motivated by an underlying assumption: using simpler, heterogenous applications yields less memory usage.
This assumption is wrong. A recent study by Lubos Lunak completely disproved it. Chief findings of the study:
- the system with heterogeneous applications (Xubuntu, promoted as the ‘lightweight’ variant of Ubuntu) did boot up into less memory usage;
- but, as soon as any useful applications were opened, Xubuntu lost the war against both KDE and GNOME
There’s a perfectly reasonable explanation to this: both GNOME and (to a larger extent) KDE share much more code than the heterogeneous applications in Xubuntu. More shared code equals a larger footprint without open applications, but an incrementally appreciable benefit when real users are actually using the machines. Especially, if I may remind you, in terminal services settings.
Low-hanging fruit
Sure. I know one type of low-hanging fruit. Compile every application and library using the gcc -Os flag. It makes smaller binaries, which are faster to read from disk, and take (a bit) less memory. The savings should add up to about 10-20 MB of RAM, according to my completely unscientific estimates.
There’s another. Disable functionality (by properly configuring the applications before compilation). I’m not sure this is the route to take.
After that, any gains become exponentially harder.
Into the future
Fortunately, all is not despair. After the study was completed, it underwent wide distribution on the Internet. That study was directly responsible for cementing the belief on shared code.
Today, both KDE and GNOME are more committed than ever on the agenda of sharing code, and optimizing this very shared code can yield significant benefits. There are big gains lurking in all that shared code, and it’s yours for the taking. It’s just a matter of applying freely available tools and the scientific method. Instead of squandering time placing decals on your hot rod, why not try some real software tuning?
As for Ubuntu Lite: I just hope the Ubuntu Lite guys realized, independently, what you just read, and surprise us with newer ideas.
January 13th, 2007 at 7:47
Lunak’s study is naturally biased, as he works for KDE. I tend to believe that great many amount of people would run OpenOffice and Firefox even under KDE and Gnome, so there is no shared code benefit.
January 13th, 2007 at 8:15
[...] Original post by Rudd-O [...]
January 13th, 2007 at 9:01
I use neither GNOME nor KDE, both too slow. Even on a fast ThinkPad-T42 I prefer ICEWM: smaller footprint and fast.
However I am forced to use Gnome libraries for gedit and gnumeric. I would also like to use kstar but that brings so much baggage that I skip it.
I am a retired big iron programmer and like to keep my my environments slim and simple and currently use Debian.
January 13th, 2007 at 15:02
I am all for lite. If you really want to help people get into computers shoot for the few year old system that wont run XP or 2K all that well if at all. I have a Celeron 450 on a Asus P2BB, with a Nidvia TNT video card and 256m of memory. It runs Zenwalk (Slackware Lite?) linux faily snappy, entirey usable. Great for the kids, retired folks, and others to do the basics on the cheap. Big pigs like Gnome and KDE would crawl on these systems. Mark
January 13th, 2007 at 15:38
Mark: both KDE and GNOME, with productivity apps open, would consume less, not more, RAM, and would be faster.
Ilja: I don’t see how the study could be biased. No one is taking the study based on faith, but math and scientific testing. If you don’t like the numbers, reproduce it instead of saying Lunak is dishonest.
Of course, OpenOffice’s the real pig. I agree with all of you from that perspective.
January 13th, 2007 at 16:47
If you want a snappy distro that runs in 128MB RAM, Slax is the one. It offers KDE. Even the server version runs fairly well in 128MB. And it can be installed to the hard drive. You can modify the basic distro easily.
For a distro that runs well in 64MB RAM, DamnSmallLinux is the one. And yes, it uses both tightly compiled apps and apps which offer reduced functionality. You can also install it to the hard drive. Then, you can download additional apps, but you may slow it down again.
January 14th, 2007 at 0:00
I’ve been watching that site, and nothing has happened or changed since Feb 2006. I’ve pretty much written them off. But, I think the people at Fluxbuntu.org might have something for low-end systems.
There are thousands of old machines out there that are still quite capable of working for someone. Organizations like FreeGeek put computers in peoples homes that might otherwise not be able to afford a new system.
Checkout the fluxbuntu distro, and I’d like to read your commentary on that.
January 14th, 2007 at 0:39
“Second, the project is motivated by an underlying assumption: using simpler, heterogenous applications yields less memory usage. This assumption is wrong. A recent study by Lubos Lunak completely disproved it. Chief findings of the study:”
Umm… No. In every single test, the author ran, XFCE and WMaker spanked KDE and Gnome in memory usage EXCEPT ONE. In the only test where KDE and Gnome wins, the author pretty much admits he is cheating by saying, “This one admitedly may look a lot like I’m trying to cheat.”
Also, in the one test (out of six) where KDE & Gnome did well, maybe you can explain to me how OpenOffice, Firefox and Thunderbird are “simpler, heterogeneous applications”? Maybe if the author of the study had compared apples to apples I might have accepted his study a bit better. Don’t even get me started about how Lubos switched the default word processor from AbiWord to OpenOffice for the only test in which his DE won.
So to sum up… IF you use ALL of the large desktop environment’s native apps and you NEVER use a third party application, you will use quite a bit more RAM than lite environments, except POSSIBLY in one situation. That situation being if you use ALL of your DE applications and the Lite user uses the HEAVIEST applications available. So, do you think that OpenOffice, Thunderbird and Firefox are all the default applications for very many Lite distros?
Do you really expect to be taken seriously with all this taken into account?
I must say your widespread statement that Lite desktop environments aren’t all that lite is NOT correct AT ALL. You have 0 data to prove, and all of the “fair” tests Lubos made point against the conclusions of this article.
It would have been interesting to see what Xfce would have looked like with Abiword + Sylpheed + Ephiphany loaded, but alas, we don’t know. It seems that Lubos was more interested in making KDE look good, than trying to show all of the DE’s in their best possible light.
January 15th, 2007 at 23:00
I read the study, too. Lubos’s key mistake is in believing that KOffice is a drop-in replacement for OpenOffice.org. It isn’t…at least, not when you need high fidelity when both reading and writing files that are in Microsoft Office’s proprietary, closed formats.
I work in a so-called “Microsoft shop”, and I have tried KOffice, hoping that it would be good enough, especially with MS Word documents, to use every day. It isn’t, unfortunately. OpenOffice.org and Evolution are the two applications that allow me to be “Microsoft Free Since 2003″ even at work. I CANNOT DO THAT WITH KOFFICE. Its MS Office file format compatibility simply isn’t good enough. OpenOffice.org’s, by contrast, is.
Therefore, Lubos’s test is merely a proof of the benefits of shared libraries and not a demonstration of a real-world office situation. Is it KOffice’s fault that Microsoft keeps their file formats so closed? Of course not; it’s Microsoft’s fault, and that’s why the widespread success of OpenDocument is so important. We, the community, need to make sure that this success actually happens, by any and all means necessary.
When my employer starts considering OpenDocument as a de facto standard, I will be among the first to cheer! Until then, OpenOffice.org is mandatory if you want to use GNU/Linux in most North American–and European–workplaces…no matter which desktop you use–KDE, GNOME, BlackBox/FluxBox, CucamongaDesktop, whatever.
December 21st, 2007 at 0:15
I’m afraid i’m going to have to agree with “thetruth” on this one. out of all the tests that were done, only one showed xfce using more memory than kde or gnome and largely the reason behind this was that xfce was tested with 3rd party apps rather than the in house software- gnome and kde were both tested with native apps. had the comparison been apples to apples rather than apples to car parts and had I actually seen this myself [not once has gnome or kde even approached the speed of xfce] I might believe it but I don’t. using exactly the same programs, xfce utterly destroys kde and gnome in every conceivable way. the problem with xfce that prevents me from switching full time is the fact that it has bugs… lots of them… that the xfce teams dont seem to have any interest in fixing… ever… quite sad really, xfce works great with memory but the project seems to be entirely uninterested in fixing some key problems with the design…